Reliability Analysis of Structural Elements with Active Learning Kriging Using a New Learning Function: KO Function

Reliability analysis is frequently used in structural engineering for the calibration of load and resistance factors, safety assessment of complex systems, and reliability-based optimization of structural systems. Crude Monte Carlo simulation (MCS) provides a complete and strong solution compared to other reliability methods since it can solve multimodal and highly nonlinear problems irrespective of the continuity of the limit state and performance functions. The number of crude MCS trials for reliability analysis can be very high which adversely impacts its efficiency, especially for complex limit state functions. Surrogate models were used with MCS to form an efficient and accurate solution for the reliability analysis. Active learning Kriging (AK) MCS is one of the most robust reliability methods that takes the advantage of surrogate modeling to decrease the cost of calculation while utilizing the accuracy of MCS. The key element in AK-MCS is the learning function which determines the degree of accuracy of the reliability results (the probability of failure and the reliability index) and governs the efficiency of the analysis. This paper investigates the validity and performance of a new learning function for AK-MCS reliability analysis, named KO learning function, using two relevant examples. Analysis results indicated that the KO function is a valid learning function for AK-MCS, and it enhances the performance of AK-MCS compared to other learning functions for the studied examples.